Академический Документы
Профессиональный Документы
Культура Документы
WP/14/52
This Working Paper should not be reported as representing the views of the IMF.
The views expressed in this Working Paper are those of the author(s) and do not necessarily
represent those of the IMF or IMF policy. Working Papers describe research in progress by the
author(s) and are published to elicit comments and to further debate.
Abstract
This paper reexamines the relationship between trade integration and business cycle
synchronization (BCS) using new value-added trade data for 63 advanced and emerging
economies during 19952012. In a panel framework, we identify a strong positive impact
of trade intensity on BCSconditional on various controls, global common shocks and
country-pair heterogeneitythat is absent when gross trade data are used. That effect is
bigger in crisis times, pointing to trade as an important crisis propagation mechanism.
Bilateral intra-industry trade and trade specialization correlation also appear to increase
co-movement, indicating that not only the intensity but also the type of trade matters.
Finally, we show that dependence on Chinese final demand in value-added terms
amplifies the international spillovers and synchronizing impact of growth shocks in China.
JEL Classification Numbers: E32; F42
Keywords: Trade, Value Added, Business Cycle Synchronization, Spillovers, Asia.
Authors E-Mail Address: RDuval@imf.org; KCheng@imf.org; KOh1@imf.org;
DSeneviratne@imf.org; rsaraf@albany.edu
2
Contents
Page
I. Introduction ............................................................................................................................3
II. Stylized Facts about Business Cycle Synchronization ..........................................................5
A. Properties of Business Cycle Synchronization in Asia .............................................5
B. The Role of Global, Regional and Sub-regional Factors ..........................................7
III. The Role of Trade in Driving BCS ....................................................................................11
A. A Birds Eye View of the Literature .......................................................................11
B. Methodology ...........................................................................................................12
C. Results .....................................................................................................................21
D. Interpretation ...........................................................................................................25
E. An Additional Model: Assessing China Spillovers .................................................26
IV. Concluding Remarks and Policy Implications ..................................................................28
References ................................................................................................................................30
Appendixes
I. Further Details about the Dynamic Factor Analysis ............................................................33
II. Further Details on the Data .................................................................................................40
III. Robustness for EconometricsUsing Four Periods .........................................................45
3
I. INTRODUCTION
Over the past two decades, trade integration has increased rapidly within the world economy,
and particularly so within Asia. Gross trade in volume terms rose at an average rate of
8 percent per annum during 19902012, twice the average pace outside Asia. In
valued-added terms, trade increased at an average annual growth rate of over 10 percent
during the same period, also double the average pace outside Asia. Not only have Asian
economies traded more with one another, they have also traded differently, becoming more
vertically integrated as a tight-knit supply-chain network across the region was formed. Have
changes in trade patterns, in particular greater trade integration, led economies to move more
in lockstep, and disproportionately so within Asia?
Theoretically, the answer to that question is a priori ambiguous, and empirically the evidence
seems to be weaker than initially thought. The relationship between trade integration and
business cycle synchronization (BCS) has been subject to extensive research, motivated in
good part by the optimum currency area literature (OCA) that was pioneered by Mundell
(1961) and McKinnon (1963) and given new impetus by Frankel and Rose (1997, 1998). A
wide range of empirical papers (e.g., Frankel and Rose, 1997, 1998; Baxter and Kouparitsas,
2005; Imbs, 2004; Inklaar and others, 2008), including in an Asian context (e.g., Kumakura,
2006; Park and Shin, 2009), have found that trade intensity increases synchronization,
although the magnitude of the impact varies across studies.
However, while existing studies have relied on a variety of approaches including
cross-section, pooled and simultaneous equation techniques, and paid a good deal of attention
to endogeneity issues, they have typically not accounted for fixed country-pair factors and
common global shocks. Yet, as stressed by Kalemli-Ozcan, Papaioannou and Peydro (2013),
controlling for both is required to address omitted variable bias and thereby identify a causal
link. Indeed they and Abiad and others (2013) find the relationship between trade integration
and BCS to be insignificant, and that between financial integration and BCS to change sign
relative to a cross-section regression, when such controls are added in a panel setup. Earlier
studies that accounted for country-pair heterogeneity also found weaker or no effects of
overall trade intensity on BCS (Calderon, Chong and Stein, 2007; Shin and Wang, 2004),
although they found the type of trade to matter.
The present paper contributes to the literature on trade integration and BCS in several ways.
First, we account systematically for country-pair heterogeneity and common global shocks
throughout the analysis. Second, and crucially, we depart from all existing studies by using
value added instead of gross trade data, building on the recent joint OECD-WTO initiative on
trade in value added. As indicated for instance by Unteroberdoerster and others (2011), and
as will be illustrated below, gross trade data misrepresent trade linkages across countries
amid increasingly important supply-chain networks across the globe. Using value-added
trade data instead will prove crucial to identifying a robust impact of trade on BCS. Third,
this paper goes beyond bilateral trade intensity and explores the BCS impact of the nature of
trade and specialization, including vertical integration, trade specialization correlation, and
4
intra-industry trade, while also controlling for other influences on BCS such as financial
integration (including bank, FDI, portfolio flows) and macro-economic policy
synchronization across countries. Finally, in separate but related analysis, we also examine
the impact of a China shock on other economies and whether and how this shock is
propagated through various trade channels.
The main findings from this paper are the following:
Consistent with results from other recent studies, BCS appears to spike across the
globe during crises. Based on the quasi-correlation indicator proposed by Abiad and
others (2013), output correlations outside of Asia peaked during the 2008/09 Global
Financial Crisis (GFC), while within Asia the biggest spikes in BCS occurred during
the Asian Crisis of the late 1990s. During normal times, BCS is typically much lower
but has nonetheless been on an upward trend over the past two decades, particularly
in Asia.
Based on a dynamic factor analysis, global factors appear to be playing a major role
in driving business cycles in Asia and elsewhere.
Using a sample of 63 advanced and emerging economies spanning the last two
decades, a macro panel analysis based on value-added trade data finds that bilateral
trade intensity has a significant, positive, and robust effect on bilateral BCS among
country pairs. Significant, though smaller effects of intra-industry trade and
correlation of trade specialization are also found.
The impact of trade integration on BCS is greater in crisis times than in normal times,
that is, trade amplifies the synchronizing impact of common shocks. This is
qualitatively consistent with the concomitant findingalready featured in that
financial integration increases BCS in crisis times, even though it typically reduces it
during normal times.
Growth spillovers from Chinaa growing source of BCS, especially within Asia
are significant, sizeable, and larger in economies that are more dependent on final
demand from China in value-added terms.
The remainder of the paper proceeds as follows: Section II presents some stylized facts about
BCS, with particular focus on Asia, including results of a factor analysis on the roles played
by global, regional, and sub-regional factors. Section III explores the role of the intensity and
nature of trade for BCS while also covering effects of financial integration. It also features
additional, related analysis of the spillovers of growth shocks in China to other economies.
Section IV concludes.
5
II. STYLIZED FACTS ABOUT BUSINESS CYCLE SYNCHRONIZATION
A. Properties of Business Cycle Synchronization in Asia
Measurement
Throughout this paper, our default measurement of BCS is the instantaneous quasicorrelation measure proposed by Abiad and others (2013):
is the quasi-correlation of real GDP growth rates of country i and j in year t,
where
denotes the output growth rate of country i in year t and; and represent the mean
and standard deviation of output growth rate of country i, respectively, during the sample
period. The growth rate is measured as the first difference of the log of real GDP.
This measure has advantages over methods commonly used by similar studies because:
First, this enables the calculation of co-movement at every point in time rather than
over an interval of time. By contrast, most of the literature measures output
co-movement between two economies by the rolling Pearson correlation of actual or
detrended growth rates between a country pair over a window period. This artificially
introduces autocorrelation of the BCS time series due to a high degree overlapping
observations throughout the sample. In addition, given that this paper uses annual
data over the past two decades,1 the rolling correlation measure would likely be
dominated by outliers during the Asian crisis and the GFC.
Second, the quasi-correlation measure retains some nice statistical properties. First, it
can be easily shown that the period mean of the measure would asymptotically
converge to the standard Pearson correlation coefficient. Second, at any point in time,
the measure is not necessarily bounded between -1 and 1. As argued by Otto and
others (2001) and Inklaar and others (2008), if the BCS measure lies between -1 and
1, the error terms in the regression explaining it are unlikely to be normally
distributed.2
We are using annual data because most of the (value added) trade variables are available at the annual or lower
frequency. We also focus on the past two decades because data on many relevant variables were not available
earlier for many emerging economies.
2
Consequently, these authors transformed the dependent BCS variable so that it is not bounded between -1
and 1.
Finally, we calculate correlations based on actual rather than detrended growth rates,
because the latter crucially depend on the choice of filtering methods and we use lowfrequency data anyway (annual data over two decades).
BCS developments
Using annual data for 63 countries, including 34 advanced economies (7 of which in Asia)
and 29 emerging economies (8 of which in Asia), developments of BCS in Asia and
elsewhere since 1990 are depicted in Figure 1:
Consistent with findings from other similar studies, BCS sharply increased in crisis
times. The largest spikes occurred around the GFC outside Asia, and around the
Asian crisis of the late 1990s within Asia (Top left panel).
During normal times, BCS has been much smaller but also shown some upward trend
since the 1990s around the globe, including in Asia. That increase in BCS has been
particularly large in Asia and Latin America, although BCS within both regions was
still considerably lower than within the euro area during the 2000s. Within Asia, BCS
appears to be particularly high among ASEAN-5 economies, which include
Indonesia, Malaysia, the Philippines, Singapore, Thailand (Top right panel). Similar
results are obtained using the standard Pearson correlation coefficients (Bottom left
panel).
Chinas output co-movements vis--vis the rest of the Asia have increased, but Asian
economies continue to co-move more with Japan and the United States (bottom right
panel). This likely reflects the continued importance of global factors in driving
business cycles across the region, as examined below.
Figure 1
Instantaneous Quasi-Correlations by Region
4.0
Full sample
ASEAN-5
5
4
1.5
1.0
0.5
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
1999
1998
1997
-1
1996
-0.5
1995
0.0
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
-0.1
1990s
Intra-Asia
ASEAN-5
2000s
China supplychain
Intra-EA 11
Intra-LATAM
0.70
1990-99
0.50
2000-12
0.30
0.10
China
ASEAN-5
Full sample
Japan
Rest of Asia
Intra-LATAM
Full sample
Intra-EA 11
Rest of Asia
China supplychain
ASEAN-5
ASEAN-5
Full sample
-0.30
ASEAN-5
-0.10
Rest of Asia
Intra-Asia
1990s
2000s
Full sample
United States
Full sample
2.0
ASEAN-5
2.5
Rest of Asia
3.0
Intra-Asia
Full sample
3.5
Euro Area
8
follow an AR(2) process. The dataset features 34 economies including 13 in Asia, 4 in South
America, 14 in Europe, and 3 in North America.
Two types of models are estimated:
Type IIIn addition to the global, regional, and country factors, business cycles in
some parts of Asia are assumed to be influenced by an additional sub-regional factor.
In the first specification, Asia is divided into two regions: China supply-chain4
(China, Taiwan POC, Thailand, Korea, Philippines, and Malaysia) vs. other Asian
economies. In the second specification, the region is divided into ASEAN-5
economies (Malaysia, Indonesia, the Philippines, Thailand, and Singapore) vs. others;
in the third specification, Asia is divided into advanced economies (Hong Kong SAR,
Japan, Korea, Singapore, and Taiwan POC) vs. others.
We use quarterly GDP, consumption, and investment from 2000Q1 to 2012Q4. The model is
estimated by Bayesian Markov Chain Monte Carlo (MCMC) approach with the algorithm
proposed by Carter and Kohn (1994).
Results5
Figure 2 shows the estimated factors in the model without sub-regional factors, with a
70-percent confidence interval, which appears to be rather narrow except for Latin America.
The factors also seem rather intuitive in explaining recent economic fluctuations. For
example, the global factor shows a steady positive value during 200307, consistent with the
steady expansion through the period. Then the factor plummets before bouncing back,
capturing the GFC and recovery in the aftermath. Meanwhile, the Asian factor shows a distinguished
peak in the aftermath of the GFC, likely representing the steep recovery amidst heavy policy
stimulus throughout the region at the time.
The China supply-chain economies are identified based on the intensity of their trade linkages (in value added
terms) with China. See below.
5
Figure 2
Global Factor
7.0
6.0
5.0
4.0
3.0
2.0
1.0
0.0
-1.0
-2.0
-3.0
-4.0
2.0
1.0
0.0
-1.0
-2.0
-3.0
-4.0
-5.0
-6.0
-7.0
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
3.0
2.0
2.0
1.0
1.0
0.0
0.0
-1.0
-1.0
-2.0
-2.0
-3.0
2001
2002
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2010
2011
2012
4.0
3.0
2000
2001
-3.0
2000
2003
2004
2005
2006
2007
2008
2009
2010
2011
-4.0
2012
2000
2001
2002
2003
2004
3.0
2.0
1.0
0.0
-1.0
-2.0
-3.0
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2005
2006
2007
2008
2009
10
Next, given that factors are orthogonal to each other in the model, we can calculate the
contribution of each factor to the total variance by decomposing the variance as follows.
Figure 3
Output Variance Decomposition by Region: Type I:
Threelayer Model
0.6
Global
Regional
Country
0.5
0.4
0.3
0.2
0.1
0.0
Asia
South America
Europe
North America
0.4
Global
Regional
Subregional
Country
ASEAN-5
non-ASEAN-5
For simplicity and focus, we mainly focus here on results for output even though the model includes
consumption and investment.
11
III. THE ROLE OF TRADE IN DRIVING BCS
A. A Birds Eye View of the Literature
In this section we begin by reviewing very briefly what economic theory postulates and what
existing empirical research finds about the BCS impact of trade integration, as well as of
financial integration and policy coordination whose effects will also be examined below.
Trade integration
Theoretically, the impact of trade on BCS is ambiguous:
On the one hand, according to traditional trade theory, openness to trade should lead
to a greater specialization across countries. To the extent this holds in practice, and
insofar as business cycles are dominated by industry-specific supply shocks, higher
trade integration should reduce BCS.
On the other hand, if the patterns of specialization and trade are dominated by
intra-industry trade, greater trade integration should be associated with a higher
degree of output co-movement in the presence of industry-specific supply shocks. If
instead demand factors are the principal drivers of business cycles, greater trade
integration should also increase BCS, regardless of whether the patterns of
specialization are dominated by inter- or intra-industry trade.
Given the ambiguity of the theory, the impact of trade integration on BCS is essentially an
empirical question. And indeed this has been a heavily researched area, with cross-sectional
regression and simultaneous equation approaches essentially finding a significant positive
impactwith some disagreement regarding its magnitudeand most recent panel regression
work controlling for country-pair fixed effects and common global shocks essentially finding
no effect (see above).
Financial integration
Theoretically, financial integration, like trade integration, has an ambiguous impact on BCS:7
On the one hand, Morgan and others (2004) developed a model in which if firms in
one country are hit by negative shocks to the value of their collateral or productivity,
then in a more financially integrated world domestic and foreign banks would
decrease lending to this country and reallocate the funds to another, thereby causing
cycles to further diverge. Likewise, in the workhorse international real business cycle
(RBC) model of Backus, Kehoe and Kydland (1992), capital will leave a country hit
by a negative productivity shock and get reallocated elsewhere under complete
For greater details, see Kalemli-Ozcan and others (2013).
12
financial markets, again amplifying divergence. Another argument is that if higher
financial integration between countries encourages them to specialize, thereby
inducing greater inter-industry trade, higher financial integration could (indirectly)
reduce BCS.
On the other hand, if negative shocks hit the banking sector, then global banks would
pull funds away from all countries across the board, thereby amplifying business
cycle co-movement.
The empirical literature is not fully settled either. Kalemli-Ozcan and others (2003) find a
significantly positive relationship between specialization and risk-sharing, consistent with a
negative impact of financial integration on BCS. By contrast, in a simultaneous equations
framework, Imbs (2006) finds a positive effect. More recent studies such as Kalemli-Ozcan,
Papaioannou and Peydro (2013) identify a strong negative effect of banking integration on
output co-movement, conditional on global shocks and country-pair heterogeneity. But there
is evidence of a positive effect on BCS of the interaction between the GFC and banking
integration, suggesting that the negative association between that form of financial
integration and output co-movement is attenuated during crisis period (Abiad and others
2013; Kalemli-Ozcan, Papaioannou and Perri, 2013).
Policy coordination
Apart from trade and financial integration, policy matters for BCS. Specifically, if two
countries synchronizeon purpose or nottheir policies by implementing expansionary or
contractionary policies at the same time, BCS between these two would be expected to rise,
all else equal. Inklaar and others (2008), using data on OECD countries, confirms that similar
monetary and fiscal policies have a strong impact on BCS. Similarly, Shin and Wang (2003),
using data for Asian countries, find that monetary policy coordination has a significant and
positive impact on BCS.
B. Methodology
Empirical strategy
Our key objective is to assess the impact of trade integration on business cycle
synchronization. Given the lessons from previous studies, the core of our empirical
framework lies on two elements which, combined together, distinguish our work from
existing literature. First, our estimation strategy relies on panel regressions with fixed effects
to account for time-invariant country-pair idiosyncratic factors and time effects to account for
global common shocks affecting countries across the board. Second, we measure trade in
value added, rather than gross, terms.
As already discussed, recent panel studies typically find a much weaker or insignificant BCS
impact of bilateral trade intensity. One explanation might be that trade does indeed have no
impact on BCS once one addresses the spurious relationship between trade integration and
BCS by conditioning on common shocks or unobserved specific country-pair fixed effects.
13
However, another explanation might simply be that gross trade data used in recentand
indeed allstudies fail to capture properly true underlying trade linkages and
interdependence across countries in a world characterized by a rapid increase in the
fragmentation of production processes and a growing share of vertical trade.
For example, as shown by Xing and Detert (2010) in the case of iPhone trade, in 2009
Chinas value added only accounted for 3.6 percent of total trade with the United States, with
the rest of the value added being reaped by Germany, Japan, Korea, the United States and
other countries via their exports to China. In this case, using gross bilateral trade data vastly
overstates Chinas trade dependence vis--vis the United States, while understating other
countries trade dependence on the U.S. tradewhich is mostly indirect, via exports to China
of components that are then assembled and re-exported by China to the United States (and
other) markets. In order to correct for this potentially serious measurement issue, we
essentially use value-added trade data throughout this paper.
Furthermore, as shown in some previous studies, not only is bilateral trade intensity
important, but so are patterns of trade and specialization. Accordingly, in addition to trade
intensity, we explore the BCS effects of vertical versus horizontal integration, intra-industry
versus inter-industry trade, and the correlation of specialization across country pairs. While
this paper is not the first one to do so, to the best of our knowledge, it is the first to compute
and test for the impact of trade in value-added terms.
Data8
We begin with four trade variables: trade intensity, vertical integration, intra-industry trade,
and bilateral correlation of specialization. Table 1 depicts the within correlation of the trade
variables with each other. Since the magnitude of within correlation of the various trade
variables is low, we will be able to introduce them in our model simultaneously without
running into multi-collinearity issues.
For a more detailed discussion of how various data and indicators are computed, see Appendix II.
14
Table 1. Trade Variables "within" Correlations
Trade
Intra- Industry
Specialization
Vertical Trade
Trade
Correlation
Integration
Trade Intensity
1.00
-0.02
-0.09
-0.25
Intra-Industry Trade
-0.02
1.00
0.06
0.02
-0.09
0.06
1.00
0.03
-0.25
0.02
0.03
1.00
Trade Intensity
Note: The within correlations between two variables xijt and yijt are calculated as the correlation between
( ,
.. ) and ( ,
.. ), where, .
, .
and ..
.
.
.
.
Trade intensity
Bilateral trade intensity is the most
frequently featured trade variable in the
literature. We follow the standard definition,
except we define it in a value-added sense
using the recent OECD-WTO database on
trade in value added in goods and services.
Specifically,
Figure 4
Median Bilateral Trade Intensity with the World
(In percent of GDP)
0.6
Asia
EM Europe
Latin America
China supply-chain
ASEAN-5
0.5
0.4
0.3
0.2
0.1
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
1999
1998
1997
1996
1995
Sources: OECD-WTO, Trade in Value Added database; and IMF staff estimates.
Note: Based on the median of trade intensity of country pairs featured in each region with the world. China supply-chain
economies include China, Korea, Malaysia, the Philippines, Taiwan Province of China, and Thailand.
2012
2011
ASEAN-5
2010
2009
2008
2007
2006
2005
China supply-chain
2004
2003
2002
2001
Latin America
2000
EM Europe
1999
Asia
1998
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
1997
0.0
1996
1995
Sources: OECD-WTO, Trade in Value Added database; and IMF staff estimates.
Note: Based on the sum of trade intensity of economies featured in each region with the world. China supply-chain
economies include China, Korea, Malaysia, the Philippines, Taiwan Province of China, and Thailand.
Figure 4 (upper panel) shows that trade openness in value-added terms has increased more in
Asia than in other emerging regions since the mid-1990s. Likewise, intra-regional trade
measured here as the median bilateral trade intensity between Asian economy pairshas
risen most rapidly within Asia (lower panel).
15
Vertical integration
The next trade variable is the vertical trade integration between two countries. A priori,
vertical trade could have a specific impact on BCS over and above that of trade intensity (for
supportive empirical evidence using country industry-level data, see di Giovanni and
Levchenko, 2010). For instance, limited or inexistent short-term substitutability of inputs
could propagate shocks along a vertical supply chain in the event of disruptions in parts of it,
as was evident in Asia in the wake of the earthquake and tsunami in Japan in 2011. In
regression analysis we measure vertical integration between two countries by the extent to
which one countrys exports in value- added terms rely on intermediate inputs from the other
country. Like trade intensity, we also define it bilaterally:9
where
denotes the vertical trade integration between countries i and j;
represents
the share in country is exports that is attributable to the (foreign) value-added content
coming from country j.
Figure 5 decomposes the value of total gross exports into its domestic and foreign
value-added components for various countries and regions. It shows that the share of foreign
value added embedded in total exports has
Figure 5. Domestic and Foreign Value Added Embedded in Exports
generally increased in Asia economies,
(In percent of total gross exports)
particularly in China and in East Asia
Domestic value added
Foreign value added
100
90
reflecting the China supply-chain
80
network. However, the extent of vertical
70
61
66
70
73 72
78
60
86
87
87
88
integration varies across the region:
90
93
50
40
specifically, as displayed in Figure 6, value
30
added coming from or going to China has
20
39
34
30
27 28
22
10
14
13
13
12
10
increased across Asian economies, while
7
0
that coming from or going to Japan has
declined. Within ASEAN-5, vertical
integration with ASEAN-5 partners
1995 2012
est.
IND
1995 2012
est.
AUS, NZL
1995 2012
est.
1995 2012
est.
JPN
CHN
Sources: OECD-WTO, Trade in Value Added database; and IMF staff estimates.
1995 2012
est.
1995 2012
est.
ASEAN-5
region
Note that vertical trade intensity could alternatively be defined as the ratio of (the sums of each countrys)
foreign value added to (the sums of) GDP, in line with the definition of trade intensity above. However, the
trade and vertical trade intensity variables would then be collinear and could not be included simultaneously in
the regressions. For this reason, controlling for trade intensity, we instead assess any additional effect of
vertical (versus non-vertical) trade through a variable that is the ratio of (the sums of) foreign value added to
(the sums of) domestic value added. That said, alternative vertical trade variables were tried in the regressions,
without any of them turning out to be statistically significant and robust.
16
Figure 6
China
China
(In percent of total foreign value added embedded in each economy's total exports)
35
1995
30
45
40
35
30
25
20
15
10
5
0
2012 est.
25
20
15
10
5
0
Sources: OECD-WTO, Trade in Value Added database; and IMF staff estimates.
2012 est.
Sources: OECD-WTO, Trade in Value Added database; and IMF staff estimates.
Japan
Japan
(In percent of total foreign value added embedded in each economy's total exports)
35
1995
30
30
2012 est.
1995
25
25
2012 est.
20
20
15
15
10
10
5
Sources: OECD-WTO, Trade in Value Added database; and IMF staff estimates.
Sources: OECD-WTO Trade in Value Added database; and IMF staff estimates.
ASEAN5
ASEAN5
(In percent of total foreign value added embedded in each economy's total exports)
20
1995
18
30
2012 est.
1995
25
16
14
2012 est.
20
12
15
10
8
10
6
4
2
0
1995
0
Philippines
Indonesia
Thailand
Malaysia
Singapore
Sources: OECD-WTO, Trade in Value Added database; and IMF staff estimates.
ASEAN5
Indonesia
Malaysia
Philippines
Thailand
Singapore
Sources: OECD-WTO ,Trade in Value Added database; and IMF staff estimates.
ASEAN-5
17
has also generally increased. Figure 7
suggests that nature of integration with
partners differs between China and Japan,
with China specializing in downstream
activities (e.g., assembling) and Japan
specializing in upstream activities
(providing various inputs).10 Finally,
Figure 8 suggests that the United States and
the European Union remain the largest final
consumers of Asias supply chain, although
the importance of final demand coming
from China has increased rapidly over the
past two decades.
Figure 7
Median Vertical Trade with China
(In percent of GDP)
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
90s
00s
90s
Asia
00s
90s
ASEAN-5
00s
90s
Rest of Asia
00s
non-Asia
Sources: OECD-WTO, Trade in Value Added database; and IMF staff estimates.
Note: Calculated as period medians of the median country pair. Upstream vertical integration of China is defined as Chinese foreign value
added content of country js exports; and downstream vertical integration is defined as the country js foreign value added content of
Chinas exports.
Intra-industry Trade
5.0
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
00s
90s
Asia
00s
90s
ASEAN-5
00s
90s
Rest of Asia
00s
non-Asia
Sources: OECD-WTO, Trade in Value added database; and IMF staff estimates.
Note: Calculated as period medians of the median country pair. Upstream vertical integration of Japan is defined as Japanese
foreign value added content of country js exports; and downstream vertical integration is defined as the country js foreign
value added content of Japan's exports.
100
90
80
AUS
NZL
JPN
CHN
KOR
TWN
MYS
10
9
THA
16
2
15
10
14 26 14
11
PHL
10
IDN
15
11
13
6
SGP
18
5
10
1
16
5
3
7
12
95
17
IND
26
17
6
6
13
Latest
10
10
95
17 15
18
Latest
11 10
24
31
23
95
15
10
21
14 23
Latest
15
95
19
19
Latest
95
11
18
0 14
5
24
22
Latest
21
17
18 23 16 31
95
27
18 19
18
Latest
9
0
95
95
30 21
95
18 17
13
20 25
Latest
19
95
15
23
25
21
17
29
Latest
25
25
95
16
10
0
20 20
18
Latest
10
19
20
95
20
23
17
20
Latest
30
14 11
20
95
40
13
Latest
50
17
17
Latest
60
Latest
70
HKG
Sources: OECD-WTO ,Trade in Value Added database; and IMF staff estimates.
10
Upstream vertical integration with China is defined as foreign value added embedded in country is exports
that comes from China. Downstream vertical integration with China is defined as foreign value added
embedded in Chinas exports that come from country i. Both upstream and downstream vertical integration
indicators are computed as a percent of country is GDP.
18
Formally, bilateral intra-industry trade can be measured by the Grubel and Lloyd (1975)
index,
, for a country pair i-j in year t:
1
,
where
(
) are the exports from (imports to) country i to (from) country j in industry
h. The higher the index value, the greater the share of intra-industry trade relative to
inter-industry trade between the two countries. It is well-known that the Grubel-Lloyd index
is sensitive to the granularity of the trade classification. Regressions were run using
alternative Standard International Trade
Figure 9. Degree of Intra-Industry Trade
Classifications (SITC), at the three- or five(Median Bilateral Grubel-Lloyd Index; ranges from 0 to 1 )
digit level and for all goods or
0.35
1990s
2000s
manufacturing goods only. Results were
0.30
very consistent and robust across these
0.25
alternative measures. In light of this, we
0.20
only show those using the SITC three-digit
0.15
classification, which is also the one used to
0.10
0.05
compute the trade specialization correlation
0.00
index described further below. Figure 9
Asia
non-Asia
shows that intra-industry trade has
moderately increased in Asia, while only
slightly increased elsewhere.
Sources: United Nations, COMTRADE database; and IMF staff estimates.
Note: Calculated as period medians of the median country pairs in each group .
such that
,
,
19
represents the trade specialization correlation between countries i and j in year
where
t,
is
the average trade specialization of country i over all n industries in year t. , ( , ) is the
measure of gross exports (imports) of
Figure 10. Correlation of Trade Specializations
country i in industry h to (from) all its
(Median Bilateral Trade Specialization Correlations)
0.09
trading partners. Figure 10 shows that trade
1990s
2000s
0.08
specialization correlation has declined
0.07
0.06
noticeably in Asia, while remaining broadly
0.05
constant elsewhere. The decline in the
0.04
specialization correlations within Asia may
0.03
reflect greater vertical specialization of
0.02
0.01
different economies along the supply chain,
0.00
which may have increased the
Asia
non-Asia
complementarityand therefore reduced
the similarityof their production
structures.
Source: United Nations, UNCTADstat database.
Note: Calculated as period medians of the median country pairs in each group .
In addition to financial variables, we also control for policy coordination variables, including
fiscal policy synchronization, monetary policy synchronization, and exchange rate policy
synchronization. The measures used for fiscal, monetary and exchange rate policies in each
economy are the structural fiscal balance purged from the cycleto focus as much as
possible on the synchronization of fiscal shocks rather than of fiscal policies, and thereby
address potential reverse causality, the real policy rate purged from the cycle, and the
rigidity of the nominal bilateral exchange rate vis--vis the other country in the pair,
respectively (see Appendix II for details).
20
The main model
We begin with a baseline model that focuses exclusively on the impact of trade and
specialization on BCS:
(1)
(2)
where FINANCE includes (all or some of the) financial integration variable and POLICY
includes policy synchronization variables.
Endogeneity
Endogeneity is a standard concern in this type of regression. Specifically, trade might be
endogenous in the sense that BCS may be driven by some omitted or unobservable variables
that are correlated with trade; or there might be reverse causality as higher BCS may induce
greater trade intensity between more synchronized countries. In this context, OLS would be
both inconsistent and biased.
We address this concern in three ways. First, the inclusion of country-pair fixed effects
addresses endogeneity problems insofar as omitted variables or the transmission mechanism
through which BCS affects trade are time-invariant, such as, for example, geographical
proximity, culture, etc. Second, the use of lagged trade intensity variables in an annual panel
should further mitigate reverse causality.11 Third, following most studies, we also resort to
instrumental variables for both trade intensity and vertical integration. Specifically, for trade
intensity, we use a set of time-varying gravity variables, comprising (see Appendix II for
11
Given that intra-industry trade and trade specialization correlation are more structural in nature and less
susceptible to reverse causality, their contemporaneous values are used.
21
details): i) the product of the real GDP of the two countries; ii) a WTO membership dummy;
iii) the degree of trade cooperation between countries; iv) a geographical distance index; and
v) the average import tariff of the two countries. For vertical trade integration, in addition to
the five variables above, we add as instruments the average import tariff on intermediate
goodswhich should correlate negatively with the bilateral intensity of intermediate goods
tradeand the difference in real per capita GDP levels between the two countrieswhich
empirically correlates well with the share of intermediate goods in their bilateral trade, as
vertical integration tends to exploit factor proportion/price differences across countries, with
the China supply-chain providing a typical illustration.
C. Results
Baseline regressionsTrade only
Table 2 presents the results of our baseline model with trade variables only (i.e., Equation 1).
Due to the presence of serial correlation, standard errors are clustered at country-pair level in
all models (including subsequent tables), to allow for autocorrelation and arbitrary
heteroskedasticity for each pair (Kalemli-Ozcan, Papaioannou and Peydro, 2013). In order to
illustrate the importance of using value-added trade data rather than gross trade data,
column 1 of Table 2 shows that trade intensity in gross terms is not significant in explaining
co-movement of business cycles once country-pair and time-fixed effects are accounted for.
By contrast, column 2 shows that trade intensity in value-added terms has a highly significant
and positive effect on BCS. While we do not report below any other regressions including
gross trade data, these consistently show an insignificant coefficient, in contrast with the
significant and robust impact of value-added trade.
Column 3 shows the instrumental variable regression for the model in column 2, confirming
that trade intensity is significant and positive. Somewhat surprisinglyinsofar as any reverse
causality would be expected to be such that higher BCS increases trade intensity but
consistent with findings in earlier studies, the instrumented regressions yield a higher
coefficient estimate than do OLS regressions. In column 4, intra-industry trade and trade
specialization correlation measures are added, and both have the expected positive
coefficients, with statistical significance at 1 the percent level. This also holds true in
column 5 when trade intensity is instrumented.
22
Table 2. Business Cycle Synchronization and Trade Integration
OLS
OLS
IV
OLS
IV
OLS
(1)
(2)
(3)
(4)
(5)
(6)
0.0488***
0.249***
0.0632***
0.295***
0.0652***
(0.0154)
(0.0738)
(0.0152)
(0.0709)
(0.0158)
0.0399
(0.0262)
Intra-industry Trade
(0.00116)
(0.00119)
(0.00120)
1.261***
1.419***
1.252***
(0.157)
(0.166)
(0.160)
-0.125***
(0.0242)
Country-fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
Year-fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
First-stage F-statistic
R-squared
Observations
49.93
49.65
0.58
0.58
0.58
0.58
0.58
0.59
18224
18619
18614
18619
18614
18606
However, we fail to find a robust specific impact of vertical trade integration over and above
that of trade intensity. As shown in column 6, when the vertical trade integration measure is
added, the former three trade variables continue to be positive and significant, but vertical
integration enters negatively. However, the latter finding is not robust across alternative
specifications, for example, it becomes positive and significant when we remove time fixed
effects from the model while all other trade variables in the model keep their sign and
significance, or it changes sign when we instrument trade intensity and vertical integration, in
ways that depend on the instruments used (results not shown, available upon request).12
Therefore we discard this variable in the remainder of the paper and leave this issue for
future work. For instance, it could well be that vertical integration creates co-movement only
by propagating up and down the international supply chain only specific shocks such as
natural disasters (in the presence of limited short-term substitutability of intermediate inputs
used in production processes). This may have been the case for instance with the earthquake
and tsunami that affected Japan in 2011, which adversely affected the industrial output of
Japans downstream trade partners in the regional supply chain. Recent IMF work (IMF,
12
We also do not report alternative regressions using alternative definitions for the vertical trade variable; none
of these turned out to show statistically significant and robust effects.
23
2013) finds some evidence that growing intermediate goods trade within the GermanCentral European supply chain (Germany, Czech Republic, Hungary, Poland and the Slovak
Republic) has increased the transmission of external shocks to this group of countries.
Augmented regressionsControlling for financial integration and policy synchronization
Next, we add financial integration and policy synchronization controls to the model
(Equation 2). As shown in Table 3, overall trade intensity stays positive and significant.
Consistent with recent literature, the coefficient on banking integration is negative significant
but here, in contrast to these papers, trade intensity remains positive and significant, in both
OLS and IV regressions (columns 12). In columns 34, we add the intra-industry trade and
trade specialization correlation indicators. While trade specialization correlation stays
significant, intra-industry trade becomes insignificant albeit remaining positive. In column 5,
we add portfolio and foreign-direct investment (FDI) integration to column 1. All three
financial variables have the expected negative sign, though the results are fairly weak,
possibly due to short sample size.
Table 3. Business Cycle Synchronization and Trade Integration-Augmented Models
OLS
IV
OLS
IV
IV
IV
IV
(1)
(2)
(3)
(4)
(5)
(6)
(7)
0.117***
0.575***
0.118***
0.566***
0.851***
0.466***
0.658***
(0.0270)
(0.0898)
(0.0270)
(0.0889)
(0.280)
(0.180)
(0.196)
Intra-industry Trade
Banking Integration
0.00202
0.00151
0.000858
(0.00177)
(0.00174)
(0.00225)
0.364*
0.433**
0.803***
(0.187)
(0.184)
(0.249)
-0.0287*
-0.0343***
-0.0282*
(0.0159)
(0.0127)
(0.0158)
Portfolio Integration
(0.0140)
(0.0125)
(0.0122)
0.0587***
0.0584***
(0.0127)
(0.0130)
0.00339**
0.00345**
(0.00149)
(0.00151)
0.136***
0.137***
(0.0168)
(0.0168)
Yes
-4.897*
(2.620)
FDI Integration
-1.338
(0.952)
Country-fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
Year-fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
Yes
38.76
13.77
10.1
11.1
First-stage F-statistic
R-squared
Observations
39.29
0.66
0.65
0.66
0.65
0.77
0.66
0.65
12186
12159
12186
12159
2860
9095
9095
24
Broadly similar results hold true when we add policy synchronization variables, as shown in
columns 6 and 7. Trade intensity and banking integration are significant while results for
other trade variables are weak. As regards the policy coordination variables themselves,
monetary policy, fiscal policy and exchange rate policy synchronization all appear to be
significant determinants of business cycle co-movements.
Assessing the impact of trade and finance on BCS during crisis vs. tranquil times
We also test whether trade integration
and banking integration have a higher
impact on BCS during crisis times than
during tranquil times. For this purpose,
we include interactions between trade
and banking integration variables, on the
one hand, and all time dummies on the
other. Results are presented in Table 4.
The first two columns are the same as in
Table 3 to facilitate comparisons. The
next two columns incorporate
interactions into the models featured in
the first two columns.
Table 4. Business Cycle Synchronization and Trade Integration Crisis vs. Non-crisis times
OLS
IV
OLS
IV
(1)
(2)
(3)
(4)
0.118***
0.566***
0.0710**
0.492***
(0.0270)
(0.0889)
(0.0300)
(0.0947)
Intra-industry Trade
Banking Integration
0.00202
0.00151
0.00261
0.00210
(0.00177)
(0.00174)
(0.00164)
(0.00173)
0.364*
0.433**
0.313*
0.274
(0.187)
(0.184)
(0.166)
(0.175)
-0.00286
(0.0158)
(0.0155)
(0.0119)
0.0527**
0.777***
(0.0266)
(0.166)
(0.0127)
The coefficient of the banking integration interaction term suggests that the sign of the effect of banking
integration on BCS changes in crisis times, becoming positive (-0.00862 + .397 = .39).
14
A more parsimonious model was also estimated featuring only the interaction between trade intensity and the
GFC (year 2009 time dummy). Results, in particular for interaction terms, were largely similar to those obtained
when the full set of time dummies is included.
25
D. Interpretation
What are the quantitative implications from the econometric results? Figure 12 provides an
illustrative interpretation of the magnitude of the estimated coefficients across various
specifications. It suggests that if a country pair moves from the 25th to the 75th percentile of
the cross-country distribution of trade
Figure 12. Illustrative Impact of Explanatory Variables
intensitywhich is equivalent to an
on Comovement
(Estimated Impact on BCS of Moving from the 25th Percentile to 75th Percentile of the
increase in trade intensity by about
Crosscountry Distribution of the Variable Considered)
0.30
0.25
12 percentage points, similar for instance to
0.20
0.15
the increase in trade intensity between the
0.10
0.05
Philippines and Thailand in the past
0.00
-0.05
12 yearsthe quasi-correlation of the
growth rates would increase by around
0.2. This impact is sizeable given the
median correlation of around 0.1 across the
sample. A similar variation of the intraindustry trade and trade specialization correlation indicators has a much smaller effect.
Finally, a similar movement along the distribution of banking integration would slightly
reduce co-movement. During crisis times, however, both trade intensity and banking
integration have a high positive impact on business cycle synchronizations in both Asia and
elsewhere. As a result, during a crisis, moving from the 25th to the 75th percentile of the
distribution of trade intensity raises the quasi-correlation by almost 0.25, compared with an
impact of 0.2 during normal times. And in the case of financial integration, crises shift its
estimated sign and turn it into a contagion channel: moving from the 25th to the 75th
percentile of the cross-country distribution of financial integration implies an increase in the
quasi-correlation by some 0.25 in crisis times, in sharp contrast to its small negative
association with BCS during tranquil times.
Bank integration
Trade intensity
Bank integration
Trade specialization
correlation
Intra-industry trade
Trade intensity
All periods
Crisis
Asia
Non-Asia
Quasi-correlation
Bank integration
Intra-industry trade
Trade specialization
correlation
Trade intensity
Quasi-correlation
Bank integration
Trade specialization
correlation
Intra-industry trade
Trade intensity
26
E. An Additional Model: Assessing China Spillovers
Setup
Chinas importance in international trade has increased rapidly in the last two decades and
numerous studies have demonstrated its profound spillovers on other countries, not least in
Asia. Here we add to this literature by studying the international spillovers of a growth shock
in China and how they may vary with the strength of bilateral trade linkages with China.
Formally, along the lines of Abiad and others (2013), we estimate the following
cross-country time-series panel regression on quarterly data:
Where is the change in the log of quarterly real GDP of country i at time t, is a
country fixed effect, shockchina,t is the China growth shock in quarter t, and
includes
controls for bilateral banking integration with China as well as global factors that affect
growth like the world oil price and global financial uncertainty (measured by the VIX).
captures trade linkages between China and country i. Alternative trade
variables are tested for, as discussed below. All are constructed at a quarterly frequency by
interpolating available end-year observations using quarterly fluctuations of bilateral gross
trade. Positive coefficients would imply that the trade variable considered is a propagation
mechanism for growth shocks originating from China.
,
Following Morgan and others (2004), the China growth shock is computed simply as the
residual growth rate that remains after removing Chinas average growth rate over the sample
period and the average growth rate of all countries during a given quarter, that is, after
estimating the panel regression:
g
where g is the quarterly growth rate of country i in year t; and are country and time
dummies, and residual growth is the growth shock for country i (and that for China for
i=China). The regression is estimated over the period 1995Q12012Q4 for the 63 countries
in the sample.
27
Results
Estimation results for the China growth spillovers specification above are presented in
Table 5.15 Trade is a significant transmission channel (column 1). As expected regarding a
growth shock originating from China, statistical significance increases when only value
added exported by country i to China is consideredthat is, ignoring the value added
exported by China to country i (column 2). Finally, also consistent with priors, the estimated
effect becomes larger when only the value added exported to China for final demand
purposesthat is, the value added that stays in Chinais considered: the impact (on
growth spillovers from China to a given country i) of 1 percentage point of GDP of value
added exported by country i to China for final demand purposes is estimated to be over twice
as large as the effect of 1 percentage point of GDP of value added exported to China for any
purposefinal demand or re-export (column 3). These results are robust to the inclusion of
bilateral banking integration with China as a control variable (results not reported).
Table 5. China's Spillover Effect Identified via Various Trade Linkages
Dependent variable: Output Growth rate
Bilateral Trade Intensity x Shock
(1)
(2)
0.101**
(3.05)
0.119**
(3.53)
Country-fixed effects
Controls for global shocks (VIX, Oil prices)
R-squared
Observations
(3)
0.148***
(3.76)
Yes
Yes
0.201
2786
Yes
Yes
0.238
3169
Yes
Yes
0.210
3181
15
All models include controls for time trend and global shock including factors such as global oil prices and the
VIX.
28
Based on column 3 in Table 5, Figure 14 suggests that Chinas growth shocks have sizeable
international spillovers. Specifically, a 1 percentage increase in Chinas growth is estimated
to raise GDP growth in the median Asian economy by over 0.3 percentage point after a year,
and in the median non-Asian economy by
Figure 14. Estimated Impact of 1 Percent Growth
Surprise in China on Partner Country Growth
about 0.15 percentage point at the same
(Median GDP growth impact after one year; in percentage points)
horizon. These estimates come fairly close
0.35
to those in Ahuja and Nabar (2012) based
0.30
0.25
on a macro panel approach, but they are
0.20
larger than these authors results using a
0.15
FAVAR approach, and also larger than the
0.10
spillovers that typically come out of global
0.05
0.00
DSGE model simulations.
Asia
(Median country pair)
non-Asia
(Median country pair)
29
Perhaps the most important implication of these results is that, all else equal, BCS among
economies would be expected to continue to rise in the future as world economic integration
increases. Trade integration would be a driving force in normal times, and an amplification
mechanism in crisis times. Any growth shocks in China would also induce more
synchronization as the importance of China as a source of final demand for the rest of the
world grows bigger. Finally, future increases in financial integrationespecially in emerging
economies where it continues to lag behind trade integration, and some catch-up could
therefore be expected could also strengthen spillovers and synchronization in crisis
periods, even though they may well reduce co-movement in tranquil times.
30
REFERENCES
Abiad, A., D. Furceri, S. Kalemli-Ozcan, and A. Pescatori, 2013, Dancing Together?
Spillovers, Common Shocks, and the Role of Financial and Trade Linkages, in
World Economic Outlook, (Washington: International Monetary Fund, October),
pp. 81111.
Ahuja, A. and M. Nabar, 2012, Investment-led Growth in China: Global Spillovers, IMF
Working Paper 12/267, (Washington: International Monetary Fund).
Backus, D., P. Kehoe and F. Kydland, 1992, International Real Business Cycles, Journal
of Political Economy, Vol. 100, pp. 745775.
Baxter, M. and M. Kouparitsas, 2005, Determinants of Business Cycle Co-movement: A
Robust Analysis, Journal of Monetary Economics, Vol. 52, pp. 113157.
Bordo, M., and T. Helbling, 2010, International Business Cycle Synchronization in
Historical Perspective, NBER Working Paper No. 16103, (Massachusetts: National
Bureau of Economic Research).
Calderon, C., A. Chong, and E. Stein, 2007, Trade Intensity and Business Cycle
Synchronization: Are Developing Countries any Different? Journal of International
Economics, Vol. 71, pp. 221.
Carter, C. K. and R. Kohn, 1994, On Gibbs Sampling for State Space Models, Biometrika,
Vol. 81, pp. 541553.
Di Giovanni, J. and A. Levchenko, 2010, Putting the Parts together: Trade, Vertical
Linkages and Business Cycle Co-movement, American Economics Journal:
Macroeconomics, Vol. 2, pp. 95124.
Frankel, J. A., and A. K. Rose, 1997, Economic Structure and the Decision to Adopt a
Common Currency, Seminar Paper No 611, The Institute for International Economic
Studies (Stockholm: Stockholm University).
, 1998, The Endogeneity of the Optimum Currency Area Criteria, Royal Economic
Society, Vol. 108(499), pp. 10091025.
Grubel, H. G. and P. J. Lloyd, 1975, Intra-industry Trade: the Theory and Measurement of
International Trade in Differentiated Products, (New York: Wiley).
Hirata, H., M. Kose, and C. Otrok, 2013 Regionalization vs. Globalization, IMF Working
Paper 13/19, (Washington: International Monetary Fund).
Imbs, J., 2004, Trade, Specialization and Synchronization, Review of Economics and
Statistics, Vol. 86, pp. 723734.
, 2006, The Real Effects of Financial Integration, Journal of International
Economics, Vol. 68, pp. 296324.
31
Inklaar, R., R. J-A. Pin, and J. Haan, 2008, Trade and Business Cycle Synchronization in
OECD countriesA re-examination, European Economic Review, Vol. 52, pp.
646-666.
International Monetary Fund, 2013, German-Central European Supply ChainCluster
Report, IMF Country Report No. 13/263 (Washington, August).
Kalemli-Ozcan, Sebnem, B.E. Sorensen, and O. Yosha, 2003, Risk Sharing and Industrial
Specialization: Regional and International Evidence, The American Economic
Review, Vol. 93, pp. 903918.
, E. Papaioannou, and F. Perri, 2013, Global Banks and Crisis Transmission,
Journal of International Economics, Vol. 89, pp. 495510.
, E. Papaioannou, and J. Peydro, 2013, Financial Regulation, Financial
Globalization, and the Synchronization of Economic Activity, Journal of Finance,
Vol. LXIII, pp. 11791228.
Kim, C-J., and C. R. Nelson, 1999, State-Space Models with Regime Switching: Classical
and Gibbs-Sampling Approaches with Applications, (Massachusetts: The MIT Press).
Kumakura, M., 2006, Trade and Business Cycle Co-movement in Asia-Pacific, Journal of
Asian Economics, Vol. 17, pp. 622645.
McKinnon, R. I., Optimum Currency Areas, The American Economic Review, Vol. 53,
pp. 717725.
Mohommad, A., O. Unteroberdoerster, and J. Vichyanond, 2011, Implications of Asias
Regional Supply Chain for Rebalancing Growth, in Regional Economic Outlook:
Asia and Pacific, (Washington: International Monetary Fund, April), pp. 4760.
Morgan, D., B. Rime, and P. Strahan, 2004, Bank Integration and State Business Cycles,
Quarterly Journal of Economics, Vol. 119, pp. 15551585.
Mundell, R. A., 1961, A Theory of Optimum Currency Areas, The American Economic
Review, Vol. 51, pp. 657665.
OECD and WTO, 2012, Trade in Value-Added: Concepts, Methodologies, and Challenges,
(OECD-WTO online document). Available via the internet: www.oecd.org/sti/ind/4
9894138.pdf.
Otto, G., G. Voss, and L. Willard, 2001, Understanding OECD Output Correlations,
Research Discussion Paper No. 200105, (Sydney: Reserve Bank of Australia).
Park, Y-C. and K. Shin, 2009, Economic Integration and Changes in the Business Cycle in
East Asia: Is the Region Decoupling from the Rest of the World, Asian Economic
Papers, Vol. 8:1, pp. 107140.
32
Shin, K., and Y. Wang, 2003, Trade Integration and Business Cycle Synchronization in East
Asia, Discussion Paper No. 574, The Institute of Social and Economic Research,
(Osaka: Osaka University).
, 2004, Trade Integration and Business Cycle Co-movements: The Case of Korea
with other Asian Countries, Japan and the World Economy, Vol. 16, pp. 213230.
Xing, Y., and N. Detert, 2010, How the iPhone widens the United States Trade Deficit with
the Peoples Republic of China, ADBI Working Paper Series No. 257, (Manila:
Asian Development Bank).
33
APPENDIX I. FURTHER DETAILS ABOUT THE DYNAMIC FACTOR ANALYSIS
Data
The source of the expenditure-side real GDP components is IMF WEO data. The database
covers the period from 2000Q1 to 2012Q4 and includes 34 countries partitioned into four
geographical regions and subgroups:
Asia: China, Hong Kong SAR, India, Indonesia, Japan, Korea, Malaysia, Philippines,
Singapore, Taiwan POC, Thailand, Australia, and New Zealand
o
China supply- chain: China, Korea, Malaysia, Philippines, Taiwan POC, and
Thailand
0,
0,
Where stands for GDP, consumption, and investment growth rates of countries, . stands
. are parameters, with the . also
for factors; and represents residuals. . , . ,
called factor loadings.
Since both factors and factor loadings have to be estimated, they are only identified by the
multiplication ( . . . Their scale and sign are not independently identified. And the scale of
a factor also depends on the variance ( ) of the factor equation. Following Hirata and
others (2013), a positive sign is imposed to the factor loading of the first country in a group.
The variance ( ) of factor equations is assigned the value estimated during the initialization
34
of parameters and factors, for which we apply principle component approach and classical
estimation.
The model is estimated by the Bayesian Markov Chain Monte Carlo (MCMC) approach with
the Carter and Kohn (1994) algorithm. Using a MCMC procedure, we can generate
parameters and factors by the following steps:
1)
2)
3)
Draw variance ( ) of the error term equation from an inverted gamma distribution
conditional on coefficients ( . , . ) and factors ( . . 1 and 1/10 are assigned for the
prior distribution parameters
and
respectively (
4)
5)
6)
In step (5), the state-space representation comprises observation and transition equations:
Observation equation:
, ~ 0,
where
is an observation of output, consumption and investment for each
economy. And
,
,
,
,
are
defined as follows:
, where
.
35
0
0
, where
0,
, where
I
I
Transition equation
, ~
where
,
,
0
0
1,2
Additional results
Figure A1.1 shows the estimated regional factors in the model with sub-regions. Even though
the sub-regional factors are mostly not significantly different from zero except for a few
years, China supply-chain and advanced economies show relatively distinguished
sub-regional factors, which imply some similarity among the member countries of each
group.
As a complementary model, we follow Bordo and Helbling (2010), who combined a VAR
approach with the baseline model. Since the model allows for interactions between economic
variables as in a usual VAR setup, we can investigate the direct impact of the regions main
economies on their neighbors. To make the model estimable on the relative short sample, we
impose some additional simplifying assumptions: the most simple VAR(1) model is applied;
only outputnot consumption or investmentis considered; no sub-regional factors are
36
Figure A1.1. Regional and Sub-regional Factors
(Median and 15 and 85-percent percentiles)
(a) Model with China Supply-chain Sub-region
Regional Factor: Asia
China Supply-chain
Sub-region
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
0.0
-0.2
-0.2
-0.4
0.8
1.0
0.6
0.8
2012
2011
2010
2009
2008
2007
2006
Other Asia
1.5
1.0
0.4
0.6
2005
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
-1.0
2002
-0.8
-1.2
2001
-0.6
-1.0
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
-1.0
-0.8
2000
-0.5
2004
-0.4
-0.6
2003
0.0
2002
0.5
2001
1.0
0.8
2000
1.5
Other Asia
0.2
0.4
0.2
0.0
0.0
-0.2
-0.2
0.5
0.0
-0.4
-0.6
-0.5
2009
2010
2011
2012
2009
2010
2011
2012
2008
2007
2006
2005
2004
2003
2002
2001
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
-1.0
2000
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
-1.0
2002
-1.0
2001
-0.8
2000
-0.8
2001
-0.6
2000
-0.4
1.0
1.0
1.0
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.2
0.2
0.0
0.0
-0.2
2008
2007
2006
2005
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
2012
2011
2010
2009
2008
2007
2006
-1.0
2005
-1.2
2004
-0.8
-0.8
2003
-1.0
2002
-0.6
-0.6
2001
-0.4
-0.8
2000
-0.6
-0.4
2004
-0.4
-0.2
2003
-0.2
2002
0.0
2001
0.2
2000
0.4
37
included. Since we add the VAR(1) term in the equation, we also drop the dynamics of the
error term. Outside of the VAR(1) term, the complementary model thus looks like a
simplified version of the baseline model:16
,
~
0,
0,
Even under the simplifying assumptions, the model has still too many explanatory variables.
Since the sample includes 34 countries, each observation equation will have 36 explanatory
variables (34 lagged variables plus two factors) and will leave only a small number of
degrees of freedom. To address this issue, Bordo and Helbling (2010) restrict directly the
by allowing non-zeros only for country is own lagged GDP growth rate and one of two
variables, namely the center country or an important trading partner. In this section, we
allow non-zero parameters for the countrys own lagged GDP growth rate, the United States,
and the largest economy in each region (China, Brazil, Germany, and Australia,
respectively). The estimation method is almost the same as for the baseline model.
Figure A1.2 shows the estimated factors of the complementary model. The global factor is
estimated tightly and is similar to that in the baseline model. The Asian regional factor is also
consistent with that in the baseline model over the second half of the period, but not clearly
distinguished otherwise. In the case of the European factor, there is larger downward trend
after the GFC than in the baseline model. Since the factor is the common trend after
considering the direct impacts from main economies, the comparison between this and the
baseline models suggest that Europe must have benefited from interactions with the United
States and German economies after the GFC.
As with the baseline model, we can decompose the unconditional variance with a little
additional complication: is a vector which consists of output ( ) of individual countries;
is a vector obtained by stacking ;
is the matrix comprised of the VAR parameter,
and is a diagonal variance matrix made of :
1
16
1 ,
Even though the baseline model allows for serial correlation in the error term ( , the error terms are
restricted to be mutually uncorrelated. So the main difference between the baseline and complementary models
is that the latter features the lagged variables on the right-hand side.
38
Figure A1.2. Global and Regional Factors in the Model with AR(1)
(Median and 15 and 85-percent percentiles)
Global Factor
1.5
2.0
2.0
1.5
1.5
1.0
1.0
0.5
1.0
0.5
0.5
0.0
0.0
0.0
-0.5
-0.5
-1.0
-1.0
-0.5
-1.5
-2.0
-1.5
-2.0
-1.0
-2.5
-2.5
0.0
-0.5
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
-2.5
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
-0.5
-1.0
2012
2011
2010
2009
2008
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
-1.5
2001
2012
2011
2010
2009
2008
2007
2006
2005
-2.0
2004
-2.5
2003
-1.5
2002
-2.0
2001
-1.0
2000
-1.5
2000
-1.0
2002
0.0
2007
-0.5
2001
0.5
0.5
0.0
2006
0.5
1.0
2005
1.0
2004
1.0
1.5
2003
1.5
2002
2.0
1.5
2001
2.0
2000
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2000
2001
-3.0
2000
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
-1.5
2000
-3.0
39
Each variance component is further divided into impacts through the countrys own lags and
through the major countries lags. The impact through the countrys own lags is calculated
by setting non-diagonal components of
to zero. Table A1.1 shows the variance
decomposition results. The transmission of the global factor has impacts on economies
through both their own lags and those of the major countries. Unlike in the baseline model,
the influence of the global factor on Asian economies is not smaller than on its impact on
European economies. By contrast, similar to the baseline model and consistent with
Figure A1.2, the explanatory power of the Asia regional factor is relatively small, consistent
with a continued key role played by global factors in driving the regional cycle. However,
this does not necessarily mean that large Asian economies like China do not have a major
influence on their neighbors, as their impact might partly be reflected in the global factors
and might also have become sizeable only in very recent years, something which cannot be
adequately captured here as our estimation and variance decomposition are performed over
the full period 200012.
Table A1.1. Variance Decomposition in the Model with AR(1)
World
Asia
Oceania
Latin America
Europe
North America
Others
Own
0.17
0.22
0.03
0.15
0.14
0.22
0.21
Global
Trasmission
0.20
0.20
0.11
0.11
0.24
0.20
0.19
Sum
0.37
0.41
0.15
0.25
0.38
0.42
0.40
Own
0.16
0.09
0.27
0.20
0.17
0.19
0.24
Regional
Trasmission
0.07
0.07
0.10
0.04
0.07
0.09
0.07
Sum
0.23
0.17
0.37
0.24
0.24
0.28
0.31
Sum
0.60
0.58
0.52
0.49
0.63
0.70
0.71
40
APPENDIX II. FURTHER DETAILS ON THE DATA
Trade integration
The main trade variables in our dataset employ trade in value-added indicators from the
second release of the OECD-WTO TiVA database (released on May 2013). Unlike gross
trade data, value-added trade data do not count products multiple times when they cross
borders repeatedly for processing purposes (OECD-WTO, 2012). The bilateral trade intensity
and vertical integration variables used throughout Section [III] are constructed using several
OECD-WTO TiVA indicators listed below, all of which cover both goods and services:
Domestic Value Added Embodied in Gross Exports (DVA): At the country level,
it can be simply defined as the sum of two contributions: i) the direct contribution
from industries that produce exported goods and services; ii) the indirect contribution
from domestic supplier industries made through domestic upstream transactions. At
the industry level, which is not used in this paper, a third componentthat cancels
out upon aggregation at the country levelis the domestic value added that was
exported in goods and services used to produce the intermediate imports of goods and
services used by the industry in question (i.e., re-imports).
Gross Exports [as estimated in OECD-WTO TiVA database] (GR): the sum of
total domestic value added and total foreign value added embodied in exports.
While the OECD-WTO TiVA database provides invaluable data on exports in value-added
terms, the database only covers years 1995, 2000, 2005, 2008, and 2009. In order to construct
an annual database covering the years 19952012, we construct the variables mentioned
above following as closely as possible the OECD-WTO methodology and concepts but using
trade data from an alternative, annual database, namely the United Nations COMTRADE.
The series we obtain can be viewed as proxies for the OECD-WTO data. We then use these
41
series profile to interpolate the OECD-WTO data between available years, and also to
extrapolate beyond 2009. We obtain full annual series over 19952012. 17
Concretely, we use UN COMTRADE gross trade data classified according to the
Harmonized System (HS) and converted to International Standard Industrial Classification
(ISIC), in line with the SNA1993 manual and the industries used in OECD-WTO TiVA
database. Using this data along with data classified in accordance with SITC, we construct
total exports, final goods exports, intermediate exports, and intermediate imports. To do so,
we incorporate classification correspondence tables developed by the United Nations
Statistics Department (i.e., HS, ISIC, CPC, SITC, and BEC correspondence tables are used).
Based on these data, the DVA series for each industry is then estimated as the sum of three
subcomponents. The domestic direct and domestic indirect components in industries are
interpolated using the growth of total exports less intermediate imports, while the
re-imported domestic component is derived using intermediate exports growth. The FVA
series is estimated by interpolating the foreign value added with intermediate imports growth.
The (OECD-WTO-consistent) gross exports series is estimated by interpolating value-added
data with total exports growth. The time series for DVA in Foreign Final Demand is
estimated by interpolating its value-added series with the final goods exports growth.
Another trade variable in our empirical analysis, namely the Grubel and Lloyd index (GL
index), relies on bilateral gross trade data from the UN COMTRADE database using SITC,
Rev.3 at a 3-digit level. The index is constructed by adjusting for the inconsistency issues of
mirror trade data due to asymmetry in reporting exports and imports in trade statistics (i.e.,
imports of country i from j usually differ from the exports reported from j to i). Our
adjustment is rather simple, where the index is constructed using both country i and country
js trade data and the average of the two values for each year is taken. As a robustness check,
we also constructed GL indices using the minimum and the maximum of the two, as well as
simply using the GL index of the high-income country within the pair as done by Calderon
and others (2007); the results remain robust in all cases.
As a robustness check, we also constructed GL indices at different levels of disaggregation
up to the 5-digit level and for different coverage of goods; results using these alternative
indices remained broadly unchanged as well. The rationale for ultimately using the 3-digit
GL index in our empirical analysis is two-fold: i) the 3-digit GL index for all goods is widely
used in trade literature, ii) it is the level of disaggregation used for another explanatory trade
variable, namely the Trade Specialization Correlation Index, which we take directly from the
United Nations Conference on Trade and Developments UNCTADstat database.
17
Robustness of the regressions using these interpolated series is validated using four-period panel regressions
(see Appendix III).
42
Financial integration
Banking integration data are based on bilateral locational banking statistics by residency
from BIS unpublished databases. Using locational data by residency is conceptually
consistent with the residency principle of national accounts and the balance of payments. We
have total bilateral external positions (both assets and liabilities separately) over 1990 2012
for BIS-reporting countries vis--vis individual partner countries. Based on these data, the
banking integration variable is defined as below:
Banking Integration: defined as in Abiad and others (2013) as the ratio of the stock
of bilateral assets and liabilities between countries i and j in year t to the sum of these
two countries external assets and liabilities vis--vis the entire world in the previous
year t-1:
where
is bilateral banking integration between countries i and j in year t,
is the stock of assets and liabilities of country is banks vis--vis country j, and
is the total stock of asset and liabilities of country i vis--vis the world in
year t-1.
Portfolio integration data are based on the bilateral portfolio investment positions (both
equity and debt securities) provided by reporting economies to the IMFs Coordinated
Portfolio Investment Survey (CPIS), in line with the residency principle of the balance of
payments. We use all available bilateral positions in the CPIS database starting as early as
2001. The portfolio integration variable is then defined as follows:
Portfolio Integration:
where
denotes the investment holdings (equity and debt securities) of country i
in country j.
FDI integration data are based on the bilateral direct investment positions provided by the
reporting economies to the IMFs Coordinated Direct Investment Survey (CDIS), in line with
the residency principle of balance of payments. The CDIS initiative, launched in late 2010,
only covers bilateral direct investment positions starting from 2009. Therefore, for the years
2000-2008, we construct the bilateral direct investment position series using data from the
now discontinued bilateral direct investment statistics from the UNCTAD and the OECD.
The FDI integration variable is then defined as follows:
FDI Integration:
43
where
where fit is the structural balance of country i in year t purged from the impact of the
cycle by regressing the structural balance on the output gap (both using IMF WEO
data), and
are, respectively, the average and standard deviation of the
structural balance of country i over the sample period.
The exchange rate rigidity variable is measured as the negative of the volatility of
monthly nominal bilateral exchange rates. Specifically, for a country pair i-j and year
t, it is defined as the standard deviation of the monthly changes in the nominal
bilateral exchange rate between i and j during year t.
Instrumental variables
As mentioned in Section III.B above, following the traditional gravity approach, we
instrument trade intensity with a geographical distance index, the degree of trade cooperation
between countries, a time-varying dummy for the membership to the WTO, the average
import tariff of the two countries, the average intermediate goods import tariff of the two
countries, and the product of their real GDPs:
Geographical distance index: following Wei (1996) and Deardoff (1998), the
geographical distance index of country i is defined as the sum of the physical
distances of country i from all its trade partners (except country j), weighted by the
44
share of trade partners in world GDP. Similarly, geographical distance index of
country j is defined as the sum of the physical distances of country j from all its trade
partners (except country i), weighted by the share of trade partners in world GDP.
Bilateral distance is the distance between the most important cities (in terms of
population) of the two countries, which is obtained from CEPIIs GeoDist database.
Import tariffs (both total and intermediate goods): data starting 1995 are obtained
from the WTO Integrated Database (IDB) and the TRANS database. We obtain tariff
data for all countries in our sample based on HS classifications, further refined in the
case of tariffs applied only to intermediate goods.
45
APPENDIX III. ROBUSTNESS FOR ECONOMETRICSUSING FOUR PERIODS
Robustness Check
In order to check the robustness of our findings on annual data and to facilitate comparisons
with previous studies, we also construct another output co-movement index using Pearson
correlations of quarterly growth rates. Four periods are considered here: 199096,
1997-2000, 2001-07 and 2008-12. Over each of these, we compute:
CORR
corr g , g
where CORR is the Pearson correlation coefficient for quarterly growth rates of countries i
and j in period , and g is the growth rate of country i in quarter t of period . The model
used for determining the effect
Table A3.1. Business Cycle Synchronization and Trade Integration of trade integration on
Regressions over Four-time Periods
business cycle is the same as
in equation (1) with t
OLS
IV
OLS
IV
replaced with . The rightDependent Variable: Correlation of
quarterly growth rates of output
(1)
(2)
(3)
(4)
hand side variables are values
averaged over period for all
Trade Intensity
0.0585**
0.271***
0.0500*
0.254***
variables, except for trade
(0.0270)
(0.0726)
(0.0270)
(0.0731)
intensity. Since the raw
Banking Integration
-0.00514 -0.00929 -0.00309 -0.00759
OECD-WTO TiVA data are
(0.0124) (0.00955) (0.0124) (0.00960)
available only for five years,
Intra-industry Trade
0.00522*** 0.00362**
we use only those here so as to
(0.00183) (0.00154)
check whether our findings are
also robust to not interpolating
Trade Specialization Correlation
0.157
0.142
(0.168)
(0.131)
the actual data. So, for the
periods 199096, 1997-2000,
Global Financial Crisis Dummy
0.389***
0.386***
0.377***
0.377***
(0.0133)
(0.0105)
(0.0143)
(0.0112)
200107 and 200812 we use
OECD-WTO trade intensity
Country-fixed effects
Yes
Yes
Yes
Yes
values for the years 1995,
Time-fixed effects
Yes
Yes
Yes
Yes
First-stage F-statistic
26.81
29.9
2000, 2005 and (an average of)
R-squared
0.63
0.60
0.64
0.60
200809, respectively.
Observations
2034
1746
2034
1746
Table A3.1 shows that these
Source: IMF staff estimates.
Standard errors, clustered at country-pair level, are given in parentheses. Global financial
four-period panel regressions
crisis represents period 2008-12.
yield results that are broadly in
* p<0.10 , ** p<0.05 , *** p<0.01
line with those based on
annual data.